The present invention generally relates to the area of powering low voltage, high current electronics such as found in the field of computing, and much of the following description is presented in that context. The invention is applicable to a wide variety of circumstances wherein a variety of power absorbing loads may abruptly change their power absorbing characteristics (that is to say, their impedance may undergo a rapid change). The invention may also be applicable if such loads are separated physically such that the voltage which may be dropped across the dynamic impedance of the power carrying conductors is a significant fraction of the voltage delivered to such loads. It may also be increasingly applicable to applications wherein design tradeoffs are forcing a steady decrease in operating voltages. Such situations may arise in telecommunications, radar systems, vehicle power systems and the like, as well as in computing systems.
The architecture of computing systems has undergone tremendous changes in the recent past, due principally to the advance of microcomputers from the original four-bit chips running at hundreds of kilohertz to the most modern 32 and 64 bit microprocessors running at hundreds of megahertz. As the chip designers push to higher and higher speeds, problems may arise which relate to thermal issues. That is, as the speed of a circuit is increased, the internal logic switches each may discharge its surrounding capacitance that much faster. Since the energy stored in that capacitance may be considered fixed (at a given voltage), as the speed is increased, that energy, which may be dissipated in the switches, may be dumped into the switch that many more times per second. Since energy per second may be defined as power, the power lost in the switches therefore increases directly with frequency.
On the other hand, the energy stored in a capacitance may increase as the square of the voltage, so a capacitor charged to two volts may store only 44% of the energy that may be stored in that same capacitor charged to three volts. For this reason, a microcomputer designed to operate at two volts will, when run at the same speed, dissipate much less power than the same microprocessor operating at three volts. There may be a tendency, therefore, to lower the operating voltage of microprocessors.
Other considerations may cause the microprocessor to exhibit a lower maximum speed if operated at a lower voltage as compared to a higher operating voltage. That is, if a circuit is operating at full speed, and the voltage on that circuit is simply reduced, the circuit may not operate properly, and the speed of the circuit (the xe2x80x9cclock speedxe2x80x9d) may have to be reduced. To maintain full speed capability and still operate at lower voltage, the circuit may have to be redesigned to a smaller physical size. For the past few years, these steps may have been considered the general course of microprocessor design. Microprocessor designers, seeking the maximum speed for their products, may expend considerable effort evaluating any number of considerations, including:
higher speed chips and potential chip value;
higher speed chips and potential heat dissipation;
potential limitations to the removal of heat;
lower voltages and the potential reduction of heat generated at a given speed; and
smaller devices and potential speed at a given voltage.
There may be many more important trade-off considerations for the designers in evaluating microprocessor design.
The evaluation of microprocessor considerations may have lead to the production of designs that operate at lower and lower voltages. Early designs may have operated at higher voltages, such as five volts, which have been subsequently reduced to current designs operating at lower voltages, such as 2.0 volts. Further reductions may occur, and future designs might be operated at 1.8, 1.5, 1.3, 1.0, and even below one volt, perhaps as low as 0.4 volts.
Meanwhile, advances in heat removal may permit processors to run at higher and higher heat dissipation levels. Early chips may have dissipated perhaps a watt; current designs may operate at the 50 watt level, and heat removal designs in the near future may be able to dissipate as much as 150 watts of power generated by the processor. Since the power dissipated may be considered proportional to the square of the operating voltage, even as the ability to remove heat is improved, lower operating voltages may still be desirable.
All of this might be viewed in the context of higher speed chips having a higher monetary value. Therefore, designers may be driven to increase the speed, potentially driving the size of the chips smaller, the voltages lower, and the power up. As may be generally known, as the voltage drops the current increases for a given power, power being defined as voltage times current. If at the same time improvements in heat removal permit higher powers, the current may increase still further. This may mean that the current rises very rapidly. Early chips may have drawn small fractions of an ampere of supply current to operate, whereas current designs may use up to 50 amperes, and future designs may use as much as 150 amperes or more.
As the speed of the processors increase, the dynamics of their power supply requirements may also increase. A processor may be drawing very little current because it is idling, and then an event may occur (such as the arrival of a piece of key data from a memory element or a signal from an outside event) which may cause the processor to suddenly start rapid computation. This may produce an abrupt change in the current drawn by the processor, which may potentially have serious electrical consequences. It may also require an unusually low output impedance or inductance.
As may be generally known, inductance is the measure of energy storage in magnetic fields. Current-carrying conductors have associated with the current a magnetic field, which represents energy storage. As it may be generally known, the energy stored in a magnetic field is half the volume integral of the square of the magnetic field. Since the field may be considered linearly related to the current in the conductor, it may be shown that the energy stored by a current-carrying conductor is proportional to half the square of the current, and the constant of proportionality may be called the xe2x80x9cinductancexe2x80x9d of the conductor. The energy stored in the system may be supplied by the source of electrical current, and for a given power source there may be a limit to the rate at which energy can be supplied, which means that the stored energy must be built up over time. Therefore, the presence of an energy storage mechanism may slow down a circuit, as the energy may be produced and metered into the magnetic field at some rate before the current can build up.
The available voltage, the inductance, and the rate of change of current in a conductor may be related by the following equation, well known to those skilled in the art:
V=L*∂I/∂t,
where L is the inductance of the conductor, and ∂I/∂t is the rate of change of current in the conductor.
This equation may be read to provide that the voltage required to produce a given current in a load on a power system increases as the time scale is reduced, and also increases as the inductance of any connection to that load is increased. In a corresponding fashion, as the speed of microprocessors may be increased, the time scale may be reduced, and as the voltage may be reduced, the equation may be read to require the inductance to be dropped proportionally.
Often, in powering semiconductor devices, a designer may not need to consider the inductance of the connections to the device, but with modern high speed circuits these considerations may force the attention to be brought to lowering the inductance of the connections. Microprocessors may currently operate at about two volts, and may tolerate a voltage transient on their supply lines of about 7%, or 140 millivolts. These same microprocessors may require that their supply current change at a rate of nearly one ampere per nanosecond, or 109 amperes/second. The above equation may be read to indicate that an inductance of 140 picohenries (1.4*10xe2x88x9210 H) may drop a voltage of 140 millivolts. To put this number in perspective, the inductance of a wire one inch in length in free space may be approximately 20,000 picohenries. While the inductance of a connection may be reduced by paralleling redundant connections, to create a connection with an inductance of 140 picohenries with conductors about a centimeter long might require nearly 100 parallel conductors.
The foregoing discussion might provide the source of low voltage physically close to the microprocessor, which in turn might provide the source of low voltage to be physically small. While it may be suggested that capacitors might be used to supply energy during the delay interval required for the current in the conductors to rise, the inductance of the connections to the capacitors may be considered limiting to this approach. The designer may be faced with placing the source of power very close to the processor to provide adequate stability to the processor""s power source under rapid changes in current draw. This requirement may become increasingly prevalent as the voltages drop and the currents increase, because the former may reduce the allowable transient size and the latter may increase the potential rate of change of current. Both factors may reduce the permissible inductance of the connection.
The foregoing remarks may not be limited in computers to the actual central microprocessor. Other elements of a modern computer, such as memory management circuits, graphic display devices, high speed input output circuitry and other such ancillary circuitry may have been increased in speed nearly as rapidly as the central processing element, wherein the same considerations would apply.
All modern electronics circuitry, including computers, may be powered by switch-mode power conversion systems. Such a system may generally be considered to convert incoming power from the utility line to the voltages and currents required by the electronic circuitry. In low power business and consumer electronics, such as desktop personal computers, the incoming power is generally supplied as an alternating voltage, generally 115 volts in the United States, and 220 volts in much of the rest of the world. The frequency of alternation may be either 50 or 60 Hertz, depending upon location. Such utility power is generally converted to low voltage steady (direct) current, or dc, and may be regulated to a few percent in order to be useful as power for the electronic circuits. A device which may perform such conversion is generally called a xe2x80x9cpower supplyxe2x80x9d. While it may possible to create a low voltage regulated dc power source using simple transformers, rectifiers, and linear regulators, such units may generally be heavy, bulky and inefficient. In these applications it may be desirable to reduce weight and size, and these approaches may be unsuitable for this reason alone. In addition, the inefficiency of linear regulators may also be unacceptable. Efficiency may be defined as the ratio of output power to input power, and a low efficiency might imply that heat is being developed in the unit which could be transferred to the environment to keep the unit cool. Generally, the lower the efficiency the more heat to be transferred, therefore a possible reason for finding an alternate approach.
For these reasons, virtually all modern electronics circuitry is powered by switch-mode conversion systems. These systems typically operate as follows. The incoming utility power is first converted to unregulated direct current by a rectifier. The rectified dc is then converted to a higher frequency, typically hundreds of kilohertz, by electronic switches. This higher frequency power is then transformed by a suitable transformer to the appropriate voltage level; this transformer also provides isolation from the utility power, required for safety reasons. The resulting isolated higher frequency power is then rectified again, and filtered into steady direct current for use by the electronics. Regulation of the output voltage is usually accomplished by control of the conduction period of the electronic switches. The resulting power conversion unit is smaller and lighter in weight than earlier approaches because the size and weight of the transformer and output filter are reduced proportionally to the increase in frequency over the basic utility power frequency. All of this is well known in the prior art.
In a complex electronic system, various voltages may be required. For example, in a computer system the peripherals (such as disk drives) may require +12 volts, some logic circuits may require +5 volts, input/output circuits may additionally require xe2x88x925 volts, memory interface and general logic may require 3.3 volts, and the central microprocessor may require 2.0 volts. The central power source (the device that is connected directly to the utility power) standards may require delivery of delivers +12, 3.3 and xc2x15 volts, and any required lower voltages may be derived from a +5 or +12 volt supply line by additional circuitry, generally known as voltage regulation modules, or VRMs, generally placed near to the circuits that require the lower voltage. These additional circuits may again convert the higher voltage supply to high frequency ac power, modifying the voltage through control of the period of the ac power, and again re-rectifying to the lower voltage dc. The VRM may take many forms, but a commonly used circuit approach may be the so-called xe2x80x9cbuck converterxe2x80x9d, which may xe2x80x9cchopxe2x80x9d the input voltage to a square wave with an average voltage equal to the required output voltage, and then may filter the square waveform to remove the alternating component, leaving the desired low voltage dc. Because the switching action may produce abrupt transitions, and for other reasons, it may be desirable to smooth such transitions, thus it may be desirable to have a relatively high input impedance or inductance. This can, of course, conflict with the previously mentioned desire for a low output impedance or inductance.
There may be several other problems with this standard approach, and one of particular relevance here may relate to the speed of response of the regulation system. A rapid change in the load impedance may cause a disturbance in the output voltage unless corrected, possibly by some control loop. This disturbance may be caused by the response of the filtering system used to remove the alternating component from the square wave output. The speed with which the control loop can respond may depend upon the characteristics of that filtering system and also upon the frequency of operation of the converter (the xe2x80x9cswitching frequencyxe2x80x9d).
One may increase this speed of response by storing less energy in the filtering system. Such a filtering system may comprise a simple series connection of an inductor and a capacitor. Storing less energy may require reducing the value of the inductance and capacitance, but may be limited in the ability to reduce these values by the potential necessity to adequately remove the ac component (called xe2x80x9cripplexe2x80x9d) generally at the output of the filter. The ripple may be reduced for a given value of inductance and capacitance by increasing the switching frequency, but this again may be limited by the ability of the electronic switches used in creating the square waveform from the dc input. Such switches may have a limited operating frequency, and may exhibit losses (known as xe2x80x9cswitching lossesxe2x80x9d) which may increase with the operating frequency.
What is needed, then, is a VRM power conversion approach or a power supply which may operate at a relatively low frequency to permit efficient operation of the electronic switches, which may have a low output ripple, low output impedance, high input impedance, which may store less energy in the output filter for a given frequency, and which may be at least as low in cost as prior art technology. Accordingly, substantial attempts such as those previously described by those skilled in the art may not have fully addressed the considerations raised. The present invention may be considered to address many of the previously mention considerations and may be considered in some aspects a development away from that which was previously known in the art.
Accordingly, it is an object of the present invention to provide a means of converting medium voltage dc power to a low voltage dc power at high current, permitting operation at higher efficiency than can be achieved using prior art techniques.
It is another object of the present invention to maintain that efficiency over a wide range of load conditions.
It is yet another object of the invention to provide a source of low voltage dc power at high currents which can sustain its voltage across a varying load even in the presence of high rates of change of current draw.
It is also an object of the preset invention to provide closer control of the output voltage of the power converter, even for extremely short time periods. That is to say, it is an object to provide a power source with better transient response to changes in load.
It is a further object of the invention to provide a power conversion system which stores less energy than that required by the prior art.
It is additionally an object of the present invention to provide a power conversion system which can be produced at lower cost than alternative approaches with similar characteristics.
Accordingly, the present invention is directed to a system of power conversion for performing a conversion from medium voltage dc to low voltage, high current dc at the point of power consumption with high efficiency and fast response.
The present invention utilizes a plurality of simple power converters, combined with coupled inductors, so arranged that the group of converters act together to produce a combined output which exhibits low voltage, high current, and fast regulation response.